JPS5911152B2 - Pattern matching method and its implementation device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention provides a pattern matching method, in particular a pattern matching method that compresses information and performs excellent pattern matching when pattern matching is performed by scanning a screen on which a plurality of pixel elements are arranged. 35 and the apparatus for implementing it.

Conventionally, in pattern recognition of characters, images, etc., a two-dimensional screen is divided into a large number of pixels in a matrix, and the electrical information corresponding to the optical image of the recognized pattern projected onto this screen is divided. A method has been proposed in which the putter 7 is recognized by extracting features of a pattern to be recognized from each pixel and finding a correlation between this and electrical information of a reference or standard pattern.
In this case, the photoelectric conversion mechanism that converts the amount of light incident on each pixel into an amount of electricity may include one using a rotating disk, one using a spot scanner, one using an image pickup tube, and one using a single row or two-dimensional arrangement of photoelectric conversion elements. Various types are known, such as those arranged in . Furthermore, scanning methods include (1) a method of sequential scanning like on a television, (2) a method of scanning photoelectric conversion elements arranged in a row in a direction perpendicular to that row, and (
3) There is a method of simultaneously extracting data in parallel from two-dimensionally arranged photoelectric conversion elements. Here, if the scanning time of each element is constant, the scanning time of all the elements will naturally increase in the order of methods (1), (2), and (3) described above.
The above-mentioned method 3>S, that is, the method of parallel to column simultaneous processing, has the shortest scanning time.

However, since data from each element must be processed simultaneously, the circuit configuration becomes complicated. Therefore, the above method (3) can be put to practical use when the number of pixels arranged in a matrix is small, such as when recognizing numbers and characters (simple characters such as kana) (referred to as OCR). However, in cases where a large number of pixel elements are required, such as in image recognition, its practical application is extremely difficult. SUMMARY OF THE INVENTION Accordingly, the main object of the present invention is to provide a pattern matching method that can be put to practical use with a simple configuration, and an apparatus for implementing this method. The pattern matching method according to the present invention is characterized by:
In a pattern matching method that scans a screen consisting of a plurality of pixel elements and recognizes a pattern on the screen, the plurality of pixel elements are divided into a plurality of pixel groups each consisting of a plurality of pixel elements. The present invention consists in that the weighted sum of the information amount of each of the plurality of pixel groups is calculated, and a column of the weighted sum of the information amount of each pixel group is created, and this is made to represent the pattern. It is something. Further, the apparatus for carrying out the pattern matching method of the present invention is characterized by having a screen on which a pattern to be recognized consisting of a plurality of pixel groups each consisting of a plurality of pixel elements is projected, and a screen that scans the screen. The present invention comprises means for obtaining a weighted sum of information for each of a plurality of pixel groups, and means for creating a column of weighted sums of information for each pixel group. Let's describe an example in detail.

FIGS. 1A and 1B each show a screen PS used in the present invention.

Each screen PS consists of a large number of photoelectric conversion elements (pixel elements). In the example shown in FIGS. 1A and 1B, the screen PS is formed by arranging eight substantially identical pixel elements vertically and six horizontally adjacent to each other in a matrix. The screen PS is divided into six pixel groups A, B, C, D, each consisting of eight pixel elements.
This is the case where it is divided into E and F. That is, 6 pixel groups A
thru|or F are eight pixel elements A1-A8, Bl-B8,
It consists of Cl-C8, Dl-D8, El-E8 and F1-F8. For example, the number "2" is projected on the screen PS shown in FIG. 1A, as shown by the cross hatch, while the number "3" is projected on the screen PS shown in FIG. 1B. .

Therefore, the examples in FIGS. 1A and B have the numbers "
This is a case where ``2'' and ``3'' are recognized respectively. Now, as shown in FIG. 1A and B, the information amount of each pixel of the screen PS on which the numbers "2" and "3" are projected is shown in binary form, and the information amount of each pixel group A to F is weighted. If the sum is obtained by an apparatus to be described later, the table shown in FIG. 2 will be obtained.

That is, the number "2" on the screen PS is {5, 1, 2, 4, 1
, 6}, the pattern is represented by a sequence of weighted sums of information; similarly, the number [3] on the screen PS is represented by {5, 1,
, 1, 2, 5}, the pattern is represented by a sequence of weighted sums of information. Similarly, project other numbers or characters onto the screen PS, create a sequence of weighted sums of information corresponding to them, and store it, project an unknown recognition pattern onto the screen, and use that information. By comparing the sequence of weighted sums of information with the sequence of weighted sums of information of already memorized patterns (standard patterns), it is possible to determine which of the already memorized patterns the unknown recognized pattern belongs to. It is something. In this case, information can be compressed compared to the conventional method, processing time can be shortened, and the circuit configuration for this can be simplified, making it easy to put it into practical use. Next, with reference to FIG. 3, an example of an apparatus for obtaining a sequence of weighted sums of the above-mentioned information will be described. In FIG. 3, reference numeral 1 denotes a television camera, for example, which targets the screen PS shown in FIG.
The optical image is obtained on the screen PS. 2 is a clock pulse generator, and when the clock pulses from this generator are supplied to the television camera 1 via the scanning circuit 3, the pixels Al,
For example, A2...F7, F8 are sequentially scanned, and the electrical output (information) from them is transmitted to each pixel Al, A2...F7, Analog memory groups AlO, BLO, etc. provided corresponding to F8
...FLO element A, l, Al2゜゜゛゜゜゜A! 8 pieces BlV゜゜゜゜゛Bl8 pieces C1 Bi゜゜゜゜゜C182D
1F゜゛゜゜゛Dl8raElV゜゛゜゜゜E18TSF1
They are respectively supplied to Bi゜゜゜゜゜F18.

On the other hand, since the output from the clock pulse generator 2 is supplied to the shift register 4, the shift register 4 is
Analog memory element A,,, Al2...Fl7
, Fl8 sequentially, these are applied to each pixel A.
1...Sequentially memorize the Jodo from F8. Furthermore,
A counter 5 is supplied with clock pulses from its generator 2 and counts the clock pulses. The output of each analog memory group AlO-F1O is supplied as a weighted sum of information to the calculation unit 6 through operational amplifiers A2O-F2O and further through analog switches A3O-F3O, and is subjected to calculation processing.

Incidentally, since this calculation section 6 has no direct relation to the present invention, a detailed explanation thereof will be omitted. The output of counter 5 is one shot multi A4O, B4O...
・Analog switch A3O, B3O via F4O...
... are supplied to each F3O and perform opening/closing control for them. That is, as shown in FIG. 1, when using a camera 1 having a screen PS as a photoelectric conversion element having six pixel groups A to F each consisting of eight pixels, the counter 5 counts eight clock pulses. One shot multi A4O every time
- Turn on analog switches A3O-F3O in sequence by looking up F4O. Therefore, the analog memory group AlO-Fl
Pixel group A of screen PS in camera 1 stored in O
-F is supplied to the calculation unit 6 as a weighted sum output. In this manner, the arithmetic unit 6 obtains a sequence of weighted sums of information as described above. Hull the rice,
The above is a case where the electrical information from each pixel element is "O" or "1". For example, when the number "2" is projected as shown in FIG. is {5,
1, 2, 4, 1, 6}.

That is, in the case of FIG. 1A, a pattern is projected on pixels A2, A3, A4, A5, and A7, and electrical information of "1" is obtained from these, respectively, and a pattern is projected on pixels Al, A6, and A8. The electrical information of [0], that is, the electrical information cannot be obtained from these, and the electrical information of 05" as a whole is obtained from the pixel group A as a weighted sum, but now,
If a pattern is projected onto pixels Al, A3, 1X4, A6, and A8, and electrical information "1" is obtained from these pixels, no pattern is projected onto other pixels A2, A5, and A7,
Even if no information is obtained from these, that is, the pattern projected onto pixel group A is different from the former, the weighted sum of electrical information obtained from pixel group A is 651, which is the same as the former.

Therefore, in such a case, pattern recognition will be performed incorrectly. However, if weighting is applied to appropriate pixel elements, for example, pixels A2, A4, A of pixel group A
6 and A8 are weighted so that the electrical information from each pixel is obtained as [0.5], and the others are set as "1" or "O" as in the previous example.

This will be explained in the case of the above-mentioned misrecognition. Now, as shown in FIG. 1A, if a pattern is projected on pixels A2, A3, A4, A5, and A7, and no pattern is projected on other pixels Al, A6, and A8, then the electricity in this case is The information is {0,0.5,1,0.5,1,0,1,0}
Therefore, the weighted sum of electrical information from pixel group A is 64". On the other hand, the pattern is projected on pixels Al, A3, A4, A, and A8, and the pattern is projected on other pixels A2, A5, and A7. If not, the electrical information obtained is {
1,0,1,0.5,0,0.5,0,0.5}, and the weighted sum is "3.5", which is different from the former.
That is, if weighting is applied to appropriate pixels as described above, it is possible to identify patterns that would otherwise be indiscernible. An example of a device that weights pixels will be described with reference to FIG.

That is, each analog memory element All, Al2...
・On the output side of Fl3, resistors AlRもA8RFBlR~B8 each have a different resistance value depending on the required weighting.
If the output of the analog memory element All...Fl8 is supplied to the arithmetic unit 6 through the resistors AlR...F, the above-mentioned result can be obtained. can be weighted. Although the above description has been made regarding the case where a screen consisting of a large number of pixel groups is sequentially scanned, the scanning need not be limited to sequential scanning, and it is of course possible to employ various other scanning methods.

Therefore, the arrangement of pixels and the configuration of groups are not necessary in accordance with sequential scanning. For example, as shown in FIG. A similar result can be obtained even if the information is grouped together as follows, and it goes without saying that in this case as well, the accuracy of pattern recognition can be improved by weighting appropriate pixel information.

In addition, even if you arbitrarily change the number of pixels composing the screen, their group configuration, scanning method, etc. according to the pattern to be recognized,
It will be clear that similar results will be obtained.

[Brief explanation of the drawing]

1A and 1B are plan views showing photoelectric conversion screens used in the present invention, FIG. 2 is a diagram showing the amount of information obtained from the screen shown in FIG. 1, and FIG. 3 is a device implementing the present invention. FIG. 4 is a plan view of another example of the photoelectric conversion screen used in the present invention.

Claims (1)

[Claims] 1. A pattern matching method that sequentially scans a screen consisting of a plurality of pixel elements arranged in the row (horizontal) and column (vertical) directions and recognizes a pattern projected on the screen. , the plurality of pixel elements are divided into a plurality of pixel groups each consisting of a plurality of consecutive pixel elements spanning at least two rows, and the last pixel of each pixel group is the first pixel of the following pixel group. All of the plurality of pixel elements that are continuous with the pixel elements and that constitute the screen are made to belong to one of the plurality of pixel groups, and a pattern that serves as a reference for the pattern to be recognized is projected onto the screen. A weighted sum of information on each of the plurality of pixel groups of the screen is determined, and a column of weighted sums of information indicating the reference pattern is created from a column of weighted sums of information on each of the pixel groups; Similarly, create a column of weighted sums of information representing the pattern to be recognized, and compare the column of weighted sums of information representing the pattern to be recognized with the column of weighted sums of information representing the reference pattern. A pattern matching method characterized in that it is determined whether the pattern to be recognized is the same as a reference pattern. 2. The pattern matching method according to claim 1, wherein a plurality of pixel elements constituting at least one pixel group of the plurality of pixel groups are weighted. 3. In a pattern matching device that sequentially scans a screen consisting of a plurality of pixel elements arranged in the row (horizontal) and column (vertical) directions and recognizes the pattern projected on the screen, each of the above screens is The above screen is formed of a plurality of pixel groups each consisting of a plurality of consecutive pixel elements extending over two rows, and the last pixel of each pixel group is continuous with the first pixel of the following pixel group. All of the plurality of pixel elements forming the . a device for creating a column of weighted sums of information representative of the reference and the pattern to be recognized from a column of weighted sums of information of each of the pixel groups; and a device for comparing a sequence of weighted sums of information representing the pattern to be recognized with a sequence of weighted sums of information representing a reference pattern to determine whether the pattern to be recognized is the same as the reference pattern. A pattern matching device featuring: 4. The pattern matching device according to claim 3, wherein at least one pixel group of the plurality of pixel groups is composed of a plurality of weighted pixel elements.